[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <0b146b52-054d-4f89-961a-65f29037e172@huawei.com>
Date: Tue, 10 May 2022 10:03:46 +0800
From: Yicong Yang <yangyicong@...wei.com>
To: Jay Zhou <jianjay.zhou@...wei.com>, <linux-pci@...r.kernel.org>,
<linux-kernel@...r.kernel.org>, <bhelgaas@...gle.com>,
<alex.williamson@...hat.com>
CC: <yangyicong@...ilicon.com>, <weidong.huang@...wei.com>
Subject: Re: [PATCH] pci: avoid dead lock between device reset and sriov
disable
On 2022/4/4 14:25, Jay Zhou wrote:
> Call trace of PF SRIOV disable:
> sriov_numvfs_store
> device_lock <----------------- (1) get the device lock
> ->sriov_configure # e.g. vfio_pci_sriov_configure
> sriov_disable
> pci_cfg_access_lock <--- (4) wait dev->block_cfg_access to be 0
>
> Call trace of PF reset:
> reset_store
> pci_reset_function
> pci_dev_lock
> pci_cfg_access_lock <----- (2) set dev->block_cfg_access = 1
> device_lock <------------- (3) want to get the device lock
>
> These two oprations would wait for each other forever if the
> code execution sequence is (1)(2)(3)(4).
>
> Let's get the device lock and then the config access lock in
> pci_dev_lock().
>
> Signed-off-by: Jay Zhou <jianjay.zhou@...wei.com>
The patch looks good to me,
Reviewed-by: Yicong Yang <yangyicong@...ilicon.com>
I met the same problem and tried to fix it in the same way. It's worth to be merged if somebody meets
the same problem again.
https://lore.kernel.org/linux-pci/1583489997-17156-1-git-send-email-yangyicong@hisilicon.com/
> ---
> drivers/pci/pci.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 9ecce435fb3f..61a6db1d21f6 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -5103,19 +5103,19 @@ static int pci_reset_bus_function(struct pci_dev *dev, bool probe)
>
> void pci_dev_lock(struct pci_dev *dev)
> {
> - pci_cfg_access_lock(dev);
> /* block PM suspend, driver probe, etc. */
> device_lock(&dev->dev);
> + pci_cfg_access_lock(dev);
> }
> EXPORT_SYMBOL_GPL(pci_dev_lock);
>
> /* Return 1 on successful lock, 0 on contention */
> int pci_dev_trylock(struct pci_dev *dev)
> {
> - if (pci_cfg_access_trylock(dev)) {
> - if (device_trylock(&dev->dev))
> + if (device_trylock(&dev->dev)) {
> + if (pci_cfg_access_trylock(dev))
> return 1;
> - pci_cfg_access_unlock(dev);
> + device_unlock(&dev->dev);
> }
>
> return 0;
> @@ -5124,8 +5124,8 @@ EXPORT_SYMBOL_GPL(pci_dev_trylock);
>
> void pci_dev_unlock(struct pci_dev *dev)
> {
> - device_unlock(&dev->dev);
> pci_cfg_access_unlock(dev);
> + device_unlock(&dev->dev);
> }
> EXPORT_SYMBOL_GPL(pci_dev_unlock);
>
>
Powered by blists - more mailing lists